COMPUTER RECONSTRUCTION AND MODELING OF THE GREAT BUDDHA STATUE IN BAMIYAN, AFGHANISTAN

Abstract

In the valley of Bamiyan, Afghanistan, almost 2000 years ago, two big standing Buddha statues were carved out of the sedimentary rock of the region. They were 53 and 35 meters high and the Great one figured as the tallest representations of a standing Buddha. In March 2001 the Taleban militia demolished the colossal statues, as they were considered an insult to Islam. After the destruction, a consortium was established to rebuild the Great Buddha of Bamiyan at original shape, size and place. We performed the computer reconstruction of the statue, which can serve as basis for the physical reconstruction. In this paper we report the results of our photogrammetric work on the Great Buddha of Bamiyan.

Full text

COMPUTER RECONSTRUCTION AND MODELING OF THE GREAT
BUDDHA STATUE IN BAMIYAN, AFGHANISTAN
A.Gruen, F.Remondino, L.Zhang
Institute of Geodesy and Photogrammetry
ETH Zurich, Switzerland
e-mail: <agruen> <fabio> <zhangl>@geod.baug.ethz.ch
ABSTRACT:
In the valley of Bamiyan, Afghanistan, ca 2000 years ago, two big standing Buddha statues were carved out of the sedimentary rock
of the region. They were 53 and 35 meters high and the Great one figured as the tallest representations of a standing Buddha. In
March 2001 the Taleban militia demolished the colossal statues, as they were considered an insult to Islam. After the destruction, a
consortium was established to rebuild the Great Buddha of Bamiyan at original shape, size and place. Our group did the computer
reconstruction of the statue, which can serve as basis for the physical reconstruction. In this paper we report the results of the 3D
reconstruction of the Great Buddha, with particular attention to the modeling and visualization of the measurements.
KEYWORDS: Cultural Heritage, Orientation, Matching, 3D Reconstruction, Surface modeling, Visualization, Photo-realism
1. INTRODUCTION
In the great valley of Bamiyan, 200 km north-east of Kabul,
Afghanistan, two big standing Buddha statues were carved out
of the sedimentary rock of the region, at 2500 meters of
altitude. The Emperor Kanishka ordered their construction
around the second century AD. Some descendants of Greek
artists who went to Afghanistan with Alexander the Great
started the construction that lasted till the fourth century AD.
The town of Bamiyan, situated in the middle of the Silk Route,
was one of the major Buddhist centres from the second century
up to the time that Islam entered the valley in the ninth century.
The larger statue was 53 metres high while the smaller Buddha
measured 35 m. They were cut from the sandstone cliffs and
they were covered with mud and straw mixture to model the
expression of the face, the hands and the folds of the robe. To
simulate these folds of the dress, cords were draped down onto
the body and were attached with wooden pegs. The lower parts
of their arms were constructed on wooden armatures while the
upper parts of the faces were made as wooden masks. The two
giants were painted in gold and other colours and they were
decorated with dazzling ornaments. They are considered the
first series of colossal cult images in Buddhist art.
The two statues were demolished on March 2001 by the
Taleban, using mortars, dynamite, anti-aircraft weapons and
rockets. The Buddhists, the world community, ONU and
UNESCO failed to convince the Taleban to leave such works of
cultural heritage.
After the destruction, a consortium was established with the
goal to rebuild the Great Buddha of Bamiyan at original shape,
size and place. This initiative is lead by the global heritage
Internet society New7Wonders [www.new7wonders.com], with
its founder Bernard Weber and the Afghanistan Institute &
Museum, Bubendorf (Switzerland), with its director Paul
Bucherer. Our group has volunteered to perform the required
computer reconstruction, which can serve as basis for the
physical reconstruction.
In this paper we present the results of the computer
reconstruction of the Great Buddha of Bamiyan, with particular
attention to the modeling and visualization of the 3D models.
The computer reconstruction is done using three different types
of imagery in parallel and performed with automatic and
manual procedures.
2. AVAILABLE IMAGES OF THE GREAT BUDDHA
Our work is based on the use of three different types of imagery
in parallel:
1. A set of images (Figure 1 - A, B, C, D) acquired from the
Internet ("Internet images");
2. A set of tourist-type images (Figure 1 - E, F, G) acquired by
Harald Baumgartner, who visited the valley of Bamiyan
between 1965 and 1969 (“Tourist images”);
3. Three metric images (Figure 2) acquired in 1970 by Prof.
Kostka, Technical University of Graz [Kostka, 1974].
We are still processing the second data set, while results of the
other two sets are already available.
Originally our interest in the reconstruction of the statue was a
purely scientific one. We planned to investigate how such an
object could be reconstructed fully automatically using just
amateur images taken from the Internet. Then, after learning
about the efforts to actually rebuild the Great Buddha we
decided to get involved in the project beyond a purely scientific
approach and to contribute as much as we could with our
technology to the success of the work.
Out of 15 images found on the Internet, four were selected for
the processing (Figure 1). All others were not suitable for
photogrammetric processing because of very low image
quality, occlusions or small image scale. The images present
differences in size and scale, they have unknown pixel size and
camera constant and most of all the different times of
acquisition; therefore some parts visible in one image are
missing in others. Also the illumination conditions are very
different and this can create problems with automatic matching
procedure
The metric images were acquired with a TAF camera
[Finsterwalder et al., 1968], a photo-theodolit camera that
acquires photos on 13x18 cm glass plates. The original photos
were scanned by Vexcel Imaging Inc with the ULTRA SCAN
5000 at a resolution of 10 micron. The final digitized images
resulted in 16930 x 12700 pixels each (Figure 2). Their
acquisition procedure is known as well as the interior
parameters of the camera. [Kostka, 1974]
International Archives of Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol. XXXIV-5/W10. International
Worksho
p
on Visualization and Animation of Realit
y
-based 3D Models, 24-28 Februar
y
2003, Taras
p
-Vul
p
era, Switzerlan
d

Figure 1: The Internet data set (A, B, C and D) used for the 3D reconstruction.
Three images of the tourist data set (E, F and G) that we are still processing.
Figure 2: The three metric images acquired in Bamiyan in 1970 by Prof. Kostka (A, B, C)
3. PHOTOGRAMMETRIC PROCESSING
The reconstruction process consists of:
• phototriangulation (calibration, orientation and bundle
adjustment),
• image coordinate measurement (automatic matching or
manual procedure),
• point cloud and surface generation,
• texture mapping and visualization.
A contour plot of the big statue, done by Prof. Kostka [Kostka,
1974], is also available (20 cm isolines, scale 1:100). From this
plot some control points could be measured and used for the
phototriangulation.
3.1 The Internet Images
The main scientific challenge here lies in the facts that no
typical photogrammetric information about these images is
available and that existing automated image analysis techniques
will most probably fail under the given circumstances. After the
establishment of an adjusted image block (Figure 3) [Gruen et
al., 2002], the 3D reconstruction of the statue was performed
with a multi-image geometrically constrained least squares
matching software package developed at our Institute [Grün et
al., 2001]. The automatic surface reconstruction works in fully
automated mode according to the following procedure:
1. Selection of one image as the master image.
2. Extraction of a very dense pattern of feature points in the
master image using the Moravec operator. The master image
is subdivided into 7 × 7 pixel image patches and within each
patch is selected the point that has the highest interest value.
3. For each feature point, using the epipolar geometry
determined in photo-triangulation, we get the approximate
matches for the following MPGC (Multi-Photo
Geometrically Constrained) matching procedure by standard
cross-correlation technique.
4. MPGC is applied for fine matching, including patch
reshaping. MPGC exploits a priori known geometric
information to constrain the solution and simultaneous use of
more than two images [Gruen, 1985; Gruen et al., 1988;
Baltsavias, 1991].
Figure 3: A view on the recovered camera poses of the
Internet images with tie and control points
The difficulties of this data set lie in the large differences
between the images, due to the different acquisition time, the
illumination conditions and the different image scales.
A point cloud of ca 6000 points is obtained. Some holes are
present in the results (Figure 6, left) because of surface changes
due to the different time of image acquisition and to the low
texture in some areas.
3.2 The metric images
Using the information in [Kostka, 1974] and some control
points measured on the contour plot, we achieved the first ap-
proximations of the exterior and interior parameters. The final
orientation of the images is achieved using a bundle adjustment
[Gruen et al., 2002]. Then automated and manual procedures
are applied to reconstruct the 3D model of the statue.
A
B
C
A
B
C
D
E
F
G

3.2.1 Automatic Measurements
The 3D model of the Buddha statue was generated with
VirtuoZo digital photogrammetric systems. The matching
method used by VirtuoZo is a global image matching technique
based on a relaxation algorithm [VirtuoZo NT, 1999]. It uses
both grid point matching and feature point matching. The
important aspect of this matching algorithm is its smoothness
constraint satisfaction procedure. With the
smoothness constraint, poor texture areas can be bridged,
assuming that the model surface varies smoothly over the image
area. Through the VirtuoZo pre-processing module, the user
can manually or semi-automatically measure some features like
ridges, edges and regions in difficult or hidden areas. These
features are used as breaklines and planar surfaces can be
interpolated, e.g. between two edges. In VirtuoZo, first the
feature point based matching method is used to compute a
relative orientation between couples of images. Then the
measured features are used to weight the smoothness
constraints while the found approximations are used in the
following global matching method [Zhang et al., 1992]. In our
application, images B and C of the metric data set were used to
reconstruct the 3D model. A regular image grid with 9 pixels
spacing was matched using a patch size of 9 × 9 pixels and 4
pyramid levels. As result, a point cloud of ca 178 000 points is
obtained (Figure 4).
Figure 4: 3D point cloud generated with automatic matching
on the metric images (ca 178 000 points)
Due to the smoothness constraint and grid-point based
matching, the very small features of the dress were filtered or
skipped. Therefore these important small features had to be
measured manually.
3.2.2 Manual Measurements
The dress of the Buddha is rich of folds, which are between 5
and 10 cm in width. Therefore only precise manual
measurements can reconstruct the exact shape and curvature of
the dress. Therefore the metric images are imported to the
VirtuoZo stereo digitize module [Virtuozo NT, 1999] and
manual stereoscopic measurements are performed. The three
stereo-models A/C, A/B and B/C (Figure 2) are set up and
points are measured along horizontal profiles of 20 cm
increment while the folds and the main edges are measured as
breaklines. With the manual measurement a point cloud of ca
76 000 points is obtained (Figure 5) and the folds on the dress
are now well visible.
Figure 5: The point cloud of the manual measurement.
The main edges and the structures of the folds, measured as
breaklines, are well visible
4. RESULTS OF THE MODELING
4.1 Internet images
For the conversion of the point cloud to a triangular surface
mesh, a 2.5D Delauney triangulation is applied. Due to some
holes in the cloud, the created mesh surface includes some big
faces. Then the model is texturized with one image of the data
set and the result is shown in Figure 6, right.
Figure 6: Point cloud obtained from the internet images
(left). Mesh surface of the Buddha (central) and 3D model
displayed in textured mode (right image)
4.2 Metric images - Automatic measurements
Due to the smoothness constraints and grid-point based
matching, the small folds on the body of the Buddha are not
correctly measured and the point cloud of the statue and
surrounding rock looks very smooth.
For the modeling, a 2.5D Delauney triangulation is performed:
without losing its topology, the 3D surface model of the
Buddha is expanded to a plane by transforming the cartesian
coordinate system to a cylinder coordinate frame. In the defined
ρθζ
cylinder frame,
ζ
is the vertical cylinder axis crossing the
model center and parallel to the original Y-axis of the cartesian
object coordinate system.
ρ
is the euclidean distance from the
surface point to the z-axis and
θ
is the angle around the z-axis.

The 2.5D triangulation was done in the
θζ
plane and the final
shaded model of the triangulated mesh is shown in Figure 7.
The model looks a bit “bumpy“. This is due to small
measurement errors and inconsistences in surface modeling.
Figure 7: The triangulated shaded model automatically
reconstructed
Then the central image of the metric data set is mapped onto the
3D geometric surface to achieve a photorealistic virtual model
(Figure 8). The lower part of the legs are not modeled because
in the used stereomodel the legs were not visible.
Figure 8: Visualization of textured 3D model generated with
automated procedures on the metric images
4.3 Metric images - Manual measurements
In the point visualization of Figure 5 it is already possible to
distinguish the shapes of the folds on the dress. This point cloud
is not dense enough (except in the area of the folds) to generate
a complete mesh with a commercial reverse engineering
software. Therefore the generation of the surface is performed
again with the 2.5D Delauney method, by dividing the
measured point cloud in different parts. A mesh for each single
point cloud is created and then all the surfaces are merged
together with Geomagic Studio [http://www.geomagic.com].
The folds of the dress are now well reconstructed and modeled,
as shown in Figure 9.
Figure 9: Visualization in wireframe mode of the 3D
structures on the central part of the dress of the Buddha
With the commercial software some editing operations of the
meshes are also performed:
- holes filling: polygon gaps are filled by constructing
triangular structures, respecting the surrounding area;
- noise reduction: spikes are removed with smooth functions;
- edges correction: faces are splitted (divided in two parts),
moved to another location or contracted;
- polygons reduction: in some areas, the number of triangles
is reduced, preserving the shape of the object.
The final 3D model, displayed in Figure 10, shows also the
reconstructed folds of the dress. Compared to Figure 7 this
represents a much better result. For photorealistic visualization,
the central image of the metric data set is mapped onto the
model, as shown in Figure 11.
Figure 10: Shaded model of the Buddha, reconstructed with
manual measurements on the three metric images

Figure 11: The 3D model of the statue created with manual
measurements on the metric images, displayed in shaded
(left) and texturized mode (right)
5. 3D MODEL VISUALIZATION
Different tools are available to display 3D models, shareware or
commercial software, with or without real-time performance,
interactive or not.
The generated model can be visualized with a software
developed at our Institute and called Disp3D. It allows the
visualization of a 3D model as point cloud, in shaded or
textured mode, as well as with interactive navigation [Gruen et
al., 2001].
One of the few portable formats to interactively display a 3D
model like the reconstructed Buddha statue is the VRML. With
free packages like Cosmo Player or Vrweb we can display and
navigate through the model or automatically fly along some
predefined paths (Figure 12).
Figure 12: Visualization of the Buddha model with Cosmo
Player plug-in in Netscape.
Computer animation software (e.g. Maya) is generally used to
create animations of 3D models. An example is presented in
http://www.photogrammetry.ethz.ch/research/bamiyan/anim/bu
ddha.mpg. They usually render the model offline, using
antialiasing functions and producing portable videos like
MPEG or AVI.
Finally, a way to display static view of 3D models is based on
anaglyph images. An anaglyph mixes into one image a
stereoscopic view using the complementarity of colours in the
RGB channels. With coloured glasses, one can then filter the
image and see the depth information of the model (Figure 13).
Figure 13: Anaglyph image of the reconstructed 3D model
6. PHYSICAL RECONSTRUCTION
The 3D computer model that we reconstructed with the manual
procedure is used for a physical reconstruction of the Great
Buddha. At the Institute of Machine Tools and Production,
ETH Zurich, R.Zanini and J.Wirth have recreated a 1:200
model statue of the Great Buddha. The point cloud of the
photogrammetric reconstruction is imported in a digitally
programmed machine tool (Starrag NF100). The machine
works on polyurethane boxes and follows milling paths
calculated directly from the point cloud. The physical model is
created in three steps: (1) a roughing path, (2) a pre-smoothing
path and (3) the final smoothing path. The time needed for
preparing the production data was about 2 hours while the
milling of the part itself was done in about 8 hours.
Figure 14: The milling machine used for the physical
reconstruction of the Bamiyan Buddha (left) and an image of
the model (right).

7. CONCLUSIONS
The computer reconstruction of the Great Buddha of Bamiyan,
Afghanistan has been performed successfully using various
digital photogrammetric techniques. We have presented here
three versions of the 3D model, based on automated point cloud
generation using four internet images, automated point cloud
generation using three metric images and manual measurements
using three metric images. While the automated matching
methods provide for dense point clouds, they fail to model the
very fine details of the statue, e.g. the folds of the robe. Also,
some other important edges are missed. Therefore, only manual
measurements allowed to generate a 3D model accurate and
complete enough to serve as the basis for a possible physical
reconstruction in situ. The problems encountered with the
orientation of amateur images and with automated matching
could be solved in an acceptable manner. The main difficulties
of this project consisted in the transition from the point cloud
(including breaklines) to a surface model which can satisfy high
modeling and visualization demands. Since automated image
matching does not take into consideration the geometrical
conditions of the object, it is very difficult to turn such more or
less randomly generated point clouds into TIN or wireframe
structures of high quality and without losing essential
information. Commercial reverse engineering software could
also not generate correct meshes (mainly because the point
cloud is not dense enough in some parts) and conventional 2.5D
Delauney triangulation was used.
When measurements are done in manual mode it is crucial for
the operator to understand the functional behaviour of the
subsequently activated 3D modeler. An on-line modeler would
be very beneficial, as during the point measurements, the
results of this modeler could be directly plotted onto the
stereomodel and the operator could control the agreement of the
on-line model with the measurements and the structure of the
object.
A web site of the work has been established on our server and is
available at
http://www.photogrammetry.ethz.ch/research/bamiyan/
with more technical details and animations.
ACKNOWLEDGEMENT
The authors would like to thank Yuguang Li for the manual
measurements on the metric images, Robert Zanini, Joachim
Wirth and the Institute of Machine Tools Production, ETH
Zurich, for the physical reconstruction of the statue at scale
1:200, Tom Bilson, Courtauld Institute of Art, London, for
some Internet images of the Bamiyan statues and all the web
sites where we found images and information on the Bamiyan
statues.
REFERENCES
Baltsavias, E., 1991: Multiphoto Geometrically Constrained
Matching. Dissertation, IGP, ETH Zürich, Mitteilungen No. 49,
221pages.
Finsterwalder, S., 1896: Zur photogrammetrischen Praxis, pp.
225-240.
Finsterwalder, S., Hofmann, W., 1968: Photogrammetrie. De
Gruyter Lehrbuch, Berlin, pp. 119-120.
Grün, A., 1985: Adaptive Least Squares Correlation: A
powerful Image Matching Technique. South Africa Journal of
Photogrammetry, Remote Sensing and Cartography, 14 (3), pp.
175-187.
Grün, A., Baltsavias, E., 1988: Geometrically Constrained
Multiphoto Matching. Photogrammetric Engineering and
Remote Sensing, Vol. 54, No. 5, pp. 633-641.
Grün, A., Zhang, L., Visnovcova, J., 2001: Automatic
Reconstruction and Visualization of a Complex Buddha Tower
of Bayon, Angkor, Cambodia. Proceedings of 21
th
Wissenschaftlich Technische Jahrestagung of Deutsche
Gesellschaft für Photogrammetrie und Fernerkundung (DGPF),
4-7 September, Konstanz, Germany, pp.289-301.
Grün, A., Remondino, F., Zhang, L., 2002: Reconstruction of
the Great Buddha of Bamiyan, Afghanistan. International
Archives of Photogrammetry and Remote Sensing, 34(5), pp.
363-368, Corfu (Greece)
Kostka, R., 1974: Die stereophotogrammetrische Aufnahme des
Grossen Buddha in Bamiyan. Afghanistan Journal, Vol.3, nr.1,
pp. 65-74.
VirtuoZo NT, 1999, Version 3.1 Manual, Supresoft Inc.
Zhang, Z., Zhang, J., Wu, X., Zhang, H., 1992: Global Image
Matching with Relaxation Method. Proceedings of the
International Colloquium on Photogrammetry, Remote Sensing
and Geographic Information Systems, 11-14 May, Wuhan,
China, pp. 175-188.